Determination of Preferential Interaction Parameters by Multicomponent Diffusion.

Applications to Poly(ethylene glycol)-Salt-Water Ternary Mixtures

Cong Tan, John G. Albright, and Onofrio Annunziata*
Department of Chemistry, Texas Christian UniVersity, Fort Worth, Texas 76129
ReceiVed: December 13, 2007; In Final Form: February 13, 2008
Poly(ethylene glycol) (PEG) is a hydrophilic nonionic polymer used in many biochemical and pharmaceutical
applications. We report the four diffusion coefficients for the PEG-KCl-water ternary system at 25 C
using precision Rayleigh interferometry. Here, the molecular weight of PEG is 20 kg mol
-1
, which is
comparable to that of proteins. The four diffusion coefficients are examined and used to determine
thermodynamic preferential interaction coefficients. We find that the PEG preferential hydration in the presence
of KCl is 1 order of magnitude larger than that previously obtained under the same conditions for lysozyme,
a protein of similar molecular weight. In correspondence, the coupled diffusion in the PEG case was greater
than that observed in the lysozyme case. We attribute this difference to the greater exposure of polymer coils
to the surrounding fluid compared to that of globular compact proteins. Moreover, we observe that the PEG
preferential hydration significantly decreases as salt concentration increases and attribute this behavior to the
polymer collapse. Finally, we have also employed the equilibrium isopiestic method to validate the accuracy
of the preferential interaction coefficients extracted from the diffusion coefficients. This experimental
comparison represents an important contribution to the relation between diffusion and equilibrium
thermodynamics.
Introduction
The chemical potential of macromolecules in aqueous solution
is perturbed by the presence of osmolytes such as salts and small
organic molecules.
1-5
The primary thermodynamic effects of
these additives on macromolecules are believed to be preferential
solvation,
1
binding,
5
and Donnan equilibrium.
6
Understanding
how the chemical potential of a macromolecule is affected by
the concentration and nature of osmolytes is important for
characterizing free-energy changes of individual macromolecules
and it is a necessary step for describing the collective behavior
of macromolecules in solutions. By use of equilibrium dialysis
7
and vapor pressure osmometry,
8
preferential interaction coef-
ficients have been determined for several protein-osmolyte
aqueous solutions.
3
These studies have been motivated by the
strong interest in protein solubility, protein unfolding, and
biochemical reactions.
1-3
Macromolecule-osmolyte thermodynamic interactions are
also central for understanding diffusive mass transport of
macromolecules.
9-13
Indeed a gradient of osmolyte concentra-
tion can generate a gradient of macromolecule chemical potential
and vice versa. This is responsible for coupled diffusion in
systems with two or more solutes. In the case of a macromol-
ecule (1)-osmolyte (2)-solvent (0) ternary mixture,
14
diffusion
is described by the extended Ficks first law:
15
Here, C
1
and C
2
are molar concentrations of the two solutes, J
1
and J
2
are the corresponding molar fluxes, and the four D
ij
s
(with i,j ) 1,2) are the multicomponent diffusion coefficients.
The main diffusion coefficients, D
11
and D
22
, describe the flux
of a solute due to its own concentration gradient, while cross-
diffusion coefficients, D
12
and D
21
, are responsible for the flux
of a solute due to the concentration gradient of the other solute.
We have previously shown that multicomponent diffusion can
be used to extract preferential interaction coefficients for
lysozyme-salt-water systems using precision Rayleigh inter-
ferometry.
6,10-13,16
Here, we report multicomponent-diffusion measurements and
preferential interactions coefficients for the poly(ethylene
glycol)-KCl-water ternary system at 25 C using precision
Rayleigh interferometry. In this work, the molecular weight of
poly(ethylene glycol) (PEG) is 20 kg mol
-1
, comparable to that
of proteins. This investigation has been motivated by the
following three reasons. First, PEG is a hydrophilic nonionic
polymer used in many applications including the partitioning
and the precipitation of biomacromolecules using PEG-salt
aqueous systems and the preparation of PEG-based materials
for drug delivery applications.
17-19
Knowledge of multicom-
ponent diffusion coefficients and thermodynamic parameters on
PEG-salt aqueous mixtures is generally important for the above
applications. However, to our knowledge, measurements of
preferential interaction coefficients for these systems have not
been reported yet and diffusion coefficients for PEG-salt
aqueous mixtures are only available for PEG oligomers.
20,21
Second, preferential interaction coefficients have usually been
determined for protein systems. A comparison between PEG-
salt and protein-salt interactions will be valuable for under-
standing differences in hydration between PEG and proteins.
Hence, we will compare our results with those previously
obtained on the lysozyme-KCl-water ternary system.
6
Third,
the determination of preferential interaction coefficients based
on multicomponent diffusion assumes the validity of the three
postulates of nonequilibrium thermodynamics.
22
Thus an ex-
* To whom correspondence should be addressed. E-mail:
O.Annunziata@tcu.edu. Phone: (817) 257-6215. Fax: (817) 257-5851.
-J
1
) D
11
C
1
+ D
12
C
2
(1a)
-J
2
) D
21
C
1
+ D
22
C
2
(1b)
4967 J. Phys. Chem. B 2008, 112, 4967-4974
10.1021/jp711725e CCC: $40.75 2008 American Chemical Society
Published on Web 04/01/2008
perimental comparison between our diffusion-based method and
equilibrium techniques represents a very important contribution
to both equilibrium and nonequilibrium thermodynamics. We
have made this comparison by measuring preferential interaction
coefficients also using the isopiestic equilibrium method.
23
Theory
We now outline the relationship between multicomponent
diffusion and the preferential interaction coefficient. The dif-
fusion coefficients in eq 1a,b can be described relative to
different reference frames:
24
in the volume-fixed frame, the
fluxes of the components of a ternary system satisfy (J
0
)
v
V
0
+
(J
1
)
V
V
1
+ (J
2
)
V
V
2
) 0; in the solvent-fixed frame, we have (J
0
)
0
) 0. Here, J
i
and Vh
i
are the molar flux and partial molar volume
of component i, respectively. The subscript V denotes the
volume-fixed frame. The subscript 0 denotes the solvent
component when appended directly to a flux and denotes the
solvent-fixed frame when appended outside the parentheses to
an already subscripted flux or diffusion coefficient. Since
concentration differences are small and volume changes on
mixing are negligible, our measurements correspond, to an
excellent approximation, to the volume-fixed frame.
24
Here, we
report isothermal diffusion coefficients for both the volume-
and solvent-fixed frames. The solvent-fixed frame coefficients,
(D
ij
)
0
, are related to the volume-fixed frame coefficients (D
ij
)
V
by the following relations:
25-27
The linear laws of nonequilibrium thermodynamics for
isothermal diffusion in terms of the Onsager diffusion coef-
ficients (ODCs), (L
ij
)
0
, are simpler in the solvent-fixed frame.
25-27
In this frame, the fluxes for a ternary system can be written as
where
i
is the chemical potential of the ith component and
Equation 4 represents the Onsager Reciprocal Relation (ORR).
28,29
The linear relations between fluxes and thermodynamic driving
forces together with the ORR are the results of the three
postulates of nonequilibrium thermodynamics.
22
We can use eqs
3a and 3b to relate the solvent-fixed diffusion coefficients and
ODCs according to
where the
ij
(c)
s are defined by
ij
(c)
(
i
/C
j
)
T,p,Ck,k*j
, where T
is the temperature and p the pressure.
25-29
The thermodynamics of macromolecule-salt aqueous mix-
tures is commonly described using molality-based chemical
potential derivatives:
ij
(m)
(
i
/m
j
)
T,p,mk,k*j
, where
i
is the
chemical potential of the ith component and
12
(m)
)
21
(m)
.
The
ij
(m)
s are related to the
ij
(c)
s by the following linear
relations:
29
The preferential interaction coefficient,
12
, is defined by:
7,14
We now show how to relate
12
to the solvent-frame diffusion
coefficients. We can first solve eq 5a-d with respect to (L
12
)
0
and (L
21
)
0
. We can then use eq 4 to obtain:
By inserting eq 6a-d into eq 8 and solving with respect to

12
(m)
/
22
(m)
, we obtain:
Equation 9 shows that ternary diffusion coefficients can be used
to determine
12
(m)
/
22
(m)
, provided that
11
(m)
and
22
(m)
are known.
We further remark that the accuracy of
11
(m)
need not be high,
i.e., 100% error in
11
(m)
results in only 1% error in
12
(m)
/
22
(m)
.
This implies that
11
(m)
) 1/m
1
is an excellent approximation for
relatively low macromolecule concentrations (10 mg/mL).
Indeed, several terms contribute only marginally to the value
of
12
(m)
/
22
(m)
and eq 9 can be approximately written as:
Materials and Methods
Materials. Poly(ethylene glycol) (PEG) with molecular
weight ca. 20 kg mol
-1
, 99% purity was purchased from Aldrich
(D
11
)
0
)
(D
11
)
V
+ [C
1
/(1 - C
1
Vh
1
- C
2
Vh
2
)][Vh
1
(D
11
)
V
+ Vh
2
(D
21
)
V
] (2a)
(D
12
)
0
)
(D
12
)
V
+ [C
1
/(1 - C
1
Vh
1
- C
2
Vh
2
)][Vh
1
(D
12
)
V
+ Vh
2
(D
22
)
V
] (2b)
(D
21
)
0
)
(D
21
)
V
+ [C
2
/(1 - C
1
Vh
1
- C
2
Vh
2
)][Vh
1
(D
11
)
V
+ Vh
2
(D
21
)
V
] (2c)
(D
22
)
0
)
(D
22
)
V
+ [C
2
/(1 - C
1
Vh
1
- C
2
Vh
2
)][Vh
1
(D
12
)
V
+ Vh
2
(D
22
)
V
] (2d)
-(J
1
)
0
) (D
11
)
0
C
1
+ (D
12
)
0
C
2
) (L
11
)
0

1
+
(L
12
)
0

2
(3a)
-(J
2
)
0
) (D
21
)
0
C
1
+ (D
22
)
0
C
2
) (L
21
)
0

1
+
(L
22
)
0

2
(3b)
(L
12
)
0
) (L
21
)
0
(4)
(D
11
)
0
) (L
11
)
0

11
(c)
+ (L
12
)
0

21
(c)
(5a)
(D
12
)
0
) (L
11
)
0

12
(c)
+ (L
12
)
0

22
(c)
(5b)
(D
21
)
0
) (L
21
)
0

11
(c)
+ (L
22
)
0

21
(c)
(5c)
(D
22
)
0
) (L
21
)
0

12
(c)
+ (L
22
)
0

22
(c)
(5d)
(C
1
/m
1
)(1 - C
1
Vh
1
- C
2
Vh
2
)
11
(c)
)
(1 - C
2
Vh
2
)
11
(m)
+ C
2
Vh
1

12
(m)
(6a)
(C
1
/m
1
)(1 - C
1
Vh
1
- C
2
Vh
2
)
12
(c)
)
(1 - C
1
Vh
1
)
12
(m)
+ C
1
Vh
2

11
(m)
(6b)
(C
2
/m
1
)(1 - C
1
Vh
1
- C
2
Vh
2
)
21
(c)
)
(1 - C
2
Vh
2
)
21
(m)
+ C
2
Vh
1

22
(m)
(6c)
(C
2
/m
1
)(1 - C
1
Vh
1
- C
2
Vh
2
)
22
(c)
)
(1 - C
1
V
1
)
22
(m)
+ C
1
Vh
2

21
(m)
(6d)

12
lim
m
1
f0
(
m
2
m
1
)
2,T,p
) -lim
m
1
f0

12
(m)

22
(m)
(7)
(D
12
)
0

11
(c)
- (D
11
)
0

12
(c)
) (D
21
)
0

22
(c)
- (D
22
)
0

21
(c)
(8)

12
(m)
/
22
(m)
) {(D
21
)
0
(1 - C
1
Vh
1
) - (D
22
)
0
C
2
Vh
1
-
(
11
(m)
/
22
(m)
)[(D
12
)
0
(1 - C
2
Vh
2
) - (D
11
)
0
C
1
Vh
2
]}/{(D
22
)
0
(1 -
C
2
Vh
2
) - (D
11
)
0
(1 - C
1
Vh
1
) + (D
12
)
0
C
2
Vh
1
- (D
21
)
0
C
1
Vh
2
}
(9)

12
(m)

22
(m)

(D
21
)
0
(D
22
)
0
- C
2
Vh
1
(10)
4968 J. Phys. Chem. B, Vol. 112, No. 16, 2008 Tan et al.
and used without further purification. Because the polydispersity
of PEG is given by a narrow Poisson distribution function, it
does not significantly affect diffusion coefficients. Thus it is
reasonable to assume that only one monodisperse macromol-
ecule is present.
30
The molecular weight, M
1
, for PEG was taken
to be 20 000 g mol
-1
and its density as 1.3 g cm
-3
for buoyancy
corrections. Deionized water was passed through a four-stage
Millipore filter system to provide high purity water for all the
experiments. The molecular weight of water, M
0
, was taken as
18.015 g cm
-3
. Mallinckrodt AR KCl with 99.9% purity was
dried by heating at 450 C for 7 h and used without further
purification. The molecular weight of KCl, M
2
, was taken to
be 74.55 g mol
-1
.
All solutions were prepared by mass with appropriate
buoyancy corrections. Stock concentrated solutions of PEG were
made by weight to an accuracy of 0.1 mg. Density measurements
(Mettler-Paar DMA40 density meter) were made on the stock
solutions for buoyancy corrections. The pairs of solutions for
each diffusion experiment were prepared by weight. For binary
PEG-water experiments, precise masses of PEG stock solutions
were diluted with pure water to reach the final target PEG
concentrations. For binary KCl-water solutions, precise masses
of pure salt were added to flasks and diluted with pure water to
reach the final target KCl concentrations. For ternary PEG-
KCl-water solutions, precise masses of PEG stock solution and
pure salt were added to flasks and diluted with pure water to
reach the final target PEG and KCl concentrations. The densities
of these solutions were measured to determine the corresponding
molar concentrations and to calculate partial molar volumes.
Diffusion Measurements. Binary and ternary mutual diffu-
sion coefficients were measured at 25.00 C with the Gosting
diffusiometer operating in the Rayleigh interferometric optical
mode.
10,11,31,32
The refractive index profile inside a diffusion
cell is measured as described in ref 11 and references therein.
We obtain 50 refractive index profiles during the course of each
experiment. Experiments were performed by the free-diffusion
method in a 10 cm vertical diffusion cell with a 2.5 cm
horizontal optical path length and a 0.3 cm width. The
temperature was regulated to (0.001 C precision and (0.01 C
accuracy. Initial step-function distributions of solute concentra-
tions were prepared with the boundary located at the center of
the cell. All experimental data were obtained before detectable
concentration changes occurred at the top and bottom ends of
the cell, consistent with the free-diffusion boundary condition.
A minimum of two experiments is required for determining the
four diffusion coefficients at a given set of mean concentrations
(Ch
1
, Ch
2
in Table 1). These two experiments must have different
combinations of solute concentration differences across the
diffusion boundary. To verify reproducibility, two other dupli-
cate experiments were performed at each set of mean concentra-
tions.
Isopiestic Measurements. To verify the accuracy of the
12
values obtained using eq 9, we have also performed isopiestic
measurements on the PEG-KCl-water ternary system. The
isopiestic method is based on the chemical equilibrium of
volatile components in a mixture.
23
For the PEG (1)-KCl (2)-
water (0) mixture, it is an excellent approximation to assume
that water is the only component that is volatile. Mixtures of
known weight and composition are placed inside open cups of
inert metal (tantalum in our case), which are then placed in a
copper block inside a desiccator. The desiccator is then partially
evacuated and placed in a thermostatted water bath at 25.00 (
0.01 C. The solutions are allowed to reach equilibrium as the
solvent redistributes itself among the mixtures. This equilibration
process normally takes a time ranging from 1 to 4 weeks. After
equilibrium has been reached, the chemical potential of the
solvent is equal for all mixtures inside the desiccator. Note that
isopiestic measurements are significantly longer than single
diffusion measurements, which, normally, take less than 6 h.
For each experiment, a given binary KCl-water solution with
known composition was prepared by weight and split into two
solutions. A known amount of solid PEG was added to one of
them. Four amounts of binary KCl-water solutions and four
amounts of ternary PEG-KCl-water solutions were placed
inside eight tantalum cups. The weights of the eight solutions
were measured by first measuring the mass of each empty cup.
Approximately 1 g of solution was placed in each cup and a
preweighed cap with an O-ring seal was immediately placed
on the cup. The assembled cap and the solution were weighed
to 0.1 mg precision, and the mass of solution calculated. The
caps were removed and the cups were placed in the apparatus
and equilibrated. When the chambers were open to remove the
cups, the set of eight caps were simultaneously pushed onto
the cups using a special cap holder. The cups were then lifted
out of the copper block and weighed to 0.1 mg precision. The
final molalities were calculated from weight changes and initial
concentrations.
For the four binary KCl-water mixtures, the average KCl
molality, m
2
, was determined. For the four ternary PEG-KCl-
water solutions, the average PEG molality, m
1
, and the average
KCl molality, m
2
, were determined. We have then computed
the quantity (m
2
- m
2
)/m
1
. Measurements at several values of
m
1
allow us to determine (m
2
/m
1
)
0
in the limit of m
1
TABLE 1: Ternary Diffusion Coefficients for the PEG-KCl-H
2
O System at 25 C
Ch1 (mM) 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
Ch2 (M) 0.2000 0.2500 0.3000 0.5000 0.9998 2.0000 2.2996
Vh0 (M
-1
) 0.01807 0.01807 0.01807 0.01806 0.01805 0.01801 0.01800
Vh1 (M
-1
) 16.76 16.77 16.76 16.79 16.84 16.92 16.95
Vh2 (M
-1
) 0.02819 0.02824 0.02836 0.02885 0.02991 0.03131 0.03164
(D11)V (10
-9
m
2
s
-1
) 0.06010 (
0.00004
0.06016 (
0.00007
0.06082 (
0.00003
0.06107 (
0.00012
0.06230 (
0.00010
0.06356 (
0.00016
0.06393 (
0.00020
(D12)V (10
-9
m
2
s
-1
) 0.000056 (
0.000001
0.000056 (
0.000002
0.000056 (
0.000001
0.000053 (
0.000003
0.000056 (
0.000002
0.000054 (
0.000002
0.000055 (
0.000003
(D21)V (10
-9
m
2
s
-1
) 14.7 ( 0.1 17.4 ( 0.2 21.2 ( 0.1 32.2 ( 0.3 61.0 ( 0.2 119.0 ( 0.9 137.9 ( 0.6
(D22)V (10
-9
m
2
s
-1
) 1.804 ( 0.001 1.809 ( 0.002 1.804 ( 0.005 1.816 ( 0.002 1.859 ( 0.001 1.963 ( 0.002 1.997 ( 0.002
(D11)0 (10
-9
m
2
s
-1
) 0.06010 0.06048 0.06121 0.06160 0.06304 0.06489 0.06537
(D12)0 (10
-9
m
2
s
-1
) 0.000070 0.000069 0.000069 0.000066 0.000070 0.000071 0.000072
(D21)0 (10
-9
m
2
s
-1
) 14.7 17.8 21.6 33.2 64.0 129.3 151.5
(D22)0 (10
-9
m
2
s
-1
) 1.804 1.822 1.820 1.843 1.919 2.097 2.162

11
(m)
/RT (Kg mol
-1
)
3949 3944 3938 3915 3857 3738 3697

22
(m)
/RT (Kg mol
-1
)
8.853 7.052 5.858 3.489 1.746 0.899 0.791
-12 5.1 ( 0.2 5.7 ( 0.2 7.1 ( 0.2 10.0 ( 0.3 17.4 ( 0.2 30.2 ( 0.6 34.3 ( 0.4
PEG-Salt Interactions by Ternary Diffusion J. Phys. Chem. B, Vol. 112, No. 16, 2008 4969
approaching zero. In this limit, we also obtain: (m
2
/m
1
)
0
)
(m
2
/m
1
)
2
according to the Gibbs-Duhem equation.
Results
In this section, we describe our results for ternary diffusion
coefficients for the PEG (1)-KCl (2)-water (0) system at
25 C. In Table 1, we report experimental parameters related
to the diffusion experiments for the ternary PEG (1)-KCl (2)-
water (0) system at seven mean KCl concentrations Ch
2
) 0.20,
0.25, 0.30, 0.50, 1.00, 2.00, and 2.30 M and the same mean
PEG concentration of Ch
1
) 0.25 mM. We observe that this
concentration falls inside the dilute-solution regime and is lower
than the PEG concentrations normally employed for the
precipitation of biomacromolecules.
17,18
The volume-frame
diffusion coefficients (D
ij
)
V
and the corresponding eigenvalues,

1
and
2
, were obtained by applying the method of the nonlinear
least-square as described ref 32. In Table 1, we also report the
partial molar volumes: Vh
1
, Vh
2
, and Vh
0
, obtained from density
measurements. The Vh
i
s were used to calculate the solvent-frame
diffusion coefficients (D
ij
)
0
from the (D
ij
)
V
s using eq 2a-d.
The chemical potential derivative,
11
(m)
, was calculated using:

11
(m)
/RT ) 1/m
1
. On the other hand, the chemical-potential
derivative,
22
(m)
, was calculated from the available activity
coefficient data of the binary KCl-water system taken at the
same m
2
values.
33
Since the experimental PEG concentration
is low, the preferential interaction coefficient,
12
, is calculated
using eq 9 with the approximation that
12
) -
12
(m)
/
22
(m)
.
13
The
calculated values of
12
are reported as a function of KCl
concentration. Experimental details related to individual ternary
diffusion experiments and binary experiments on the KCl-water
system are available as Supporting Information.
In Figure 1a-d, we illustrate the behavior of the four (D
ij
)
V
as a function of salt concentration. In Figure 1a, we show that
(D
11
)
V
increases with C
2
. This implies that PEG diffusion is
enhanced in the presence of salt. At low concentration of neutral
macromolecules, we expect that (D
11
)
V
D
m
,
34,35
where D
m
is
the infinite dilution tracer diffusion coefficient of the macro-
molecule. Indeed our experimental value for the binary PEG-
water system at C
1
) 0.25 mM is (D
11
)
v
) 0.05990 10
-9
m
2
s
-1
. This value is 4% higher than the corresponding tracer
diffusion coefficient (D
m
) 0.0574 10
-9
m
2
s
-1
).
36
According
to the Stokes-Einstein equation,
35
the isothermal dependence
of D
m
on salt concentration can be attributed to changes in salt
water viscosity and macromolecule hydrodynamic radius. To
remove the effect of viscosity, we consider the product: (D
11
)
V
-
(/
0
), where is the viscosity of the corresponding binary
KCl-water system
37
and
0
is the viscosity of water. In Figure
2, we plot (D
11
)
V
(/
0
) as a function of C
2
. This plot shows
that (D
11
)
V
(/
0
) increases linearly with salt concentration. At
the highest experimental salt concentration, its value is 8.2%
higher than that at C
2
) 0. The observed behavior is consistent
with a corresponding decrease of the PEG hydrodynamic radius.
Figure 1. Volume-fixed diffusion coefficients as functions of KCl concentration, C2, for the PEG-KCl-H2O system at C1 ) 0.25 mM and 25 C:
(a) (D11)V; (b) (D22)V; (c) (D12)V; (d) (D21)V. The solid curves are fits through the ternary experimental points. The dashed curve (data points
omitted) is a fit through the binary KCl diffusion coefficients taken from ref 33. The open circles are our measured binary values.
Figure 2. Viscosity-corrected (D11)V(/0) as a function of salt
concentration, C2, for the PEG-KCl-H2O system. The solid line is a
linear fit through the experimental points. The viscosity coefficients
of the binary salt-water systems were taken from ref 37.
4970 J. Phys. Chem. B, Vol. 112, No. 16, 2008 Tan et al.
This polymer collapse can be related to a decrease of solvent
quality in the presence of salt, consistent with the known
capability of salts to enhance liquid-liquid-phase separation
of PEG aqueous solutions.
38
As we shall discuss later, the
dependence of the preferential interaction coefficient on salt
concentration is also consistent with the observed behavior of
(D
11
)
V
(/
0
).
In Figure 1b, we show the behavior of the salt main diffusion
coefficient (D
22
)
V
(solid circles). In the same figure, we also
include the corresponding binary diffusion coefficients (open
circles) and the available literature data (dashed curve).
33
We
can see that our binary measurements are in very good
agreement with literature data. Moreover, our ternary data show
the same behavior as the corresponding binary curve. This is
strong evidence that using binary thermodynamic data to
calculate ternary
22
(m)
values is a very good approximation.
6
We can relate our ternary data to the corresponding binary data
by (D
22
)
V
) (1 - k)(D
2
)
V
, where ) C
1
Vh
1
is the polymer
volume fraction and k is a constant factor that characterizes a
small obstruction effect of the PEG macromolecules on the
motion of the small salt ions. It has been theoretically shown
that k ) 1.5 for small solutes in the presence of hard spheres.
39
For aqueous KCl in the presence of PEG, we obtain k ) 4.0 (
0.2 by fitting the experimental (D
22
)
V
using the available
expression of (D
2
)
V
and our experimental PEG volume fraction,
) 0.0042. For comparison, we have also calculated k for
aqueous KCl in the presence of lysozyme (at ) 0.0061).
6
Here, we obtain k ) 1.8 (0.1, which is close to the hard-sphere
case. We therefore conclude that the obstruction effect of PEG
on the salt ions is 2-fold higher than that of lysozyme. This
result is consistent with polymer coils being significantly more
expanded than globular compact proteins.
In Figure 1c,d, we show the behavior of the two cross-
diffusion coefficients (D
12
)
V
and (D
21
)
V
. We can see that (D
12
)
V
is positive and approximately independent of C
2
within the
experimental error. A positive value of (D
12
)
V
implies that at
uniform PEG concentration the macromolecules diffuse from
high to low salt concentration. In Figure 1d, we show that (D
21
)
V
is positive and increases with C
2
. Note that the value of (D
21
)
V
must approach zero at C
2
) 0. This is because the salt flux
becomes zero, independent of PEG concentration gradient. A
positive value of (D
21
)
V
implies that salt diffuses from high to
low PEG concentration in the presence of uniform salt concen-
tration. The behavior of both cross-diffusion coefficients is
consistent with the presence of net repulsive interactions between
PEG and KCl in water. We will further examine the cross-
diffusion coefficients in the Discussion section.
To validate the accuracy of the
12
values reported in Table
1, we have also used the isopiestic method for the determination
of the preferential interaction coefficients. Typically, the isopi-
estic method has been employed for the determination of activity
coefficients of salt-water mixtures.
23
To our knowledge, this
is the first time that this method is used for the determination
of preferential interaction coefficients of macromolecules. In
Figure 3, measurements of (m
2
/m
1
)
0)0
are shown for two
chosen KCl concentrations: C
2
) 0.25 and 1.0 M. These two
sets of data were fitted using the linear weighed least-square
method. The corresponding intercepts are the values of -
12
(see Materials and Methods). We obtain: -
12
) 5.2 ( 0.4 at
C
2
) 0.25 M and -
12
) 16.6 ( 0.7 at C
2
) 1.0 M. Here the
reported errors are standard deviations. The corresponding values
extracted from the ternary diffusion coefficients are: -
12
)
5.7 (0.2 and -
12
) 17.4 (0.2 (see Table 1). We therefore
conclude that diffusion and isopiestic measurements are in very
good agreement within the experimental error.
Discussion
In this section, we will first discuss the preferential interaction
coefficients reported in Table 1 and compare these values with
those previously obtained for the lysozyme-KCl-water system
at pH 4.5.
6
We will then examine the more complex behavior
of the cross-diffusion coefficients in relation to the system
thermodynamic behavior.
Preferential Interaction Coefficients. In Figure 4, we plot
-
12
as a function of salt molality m
2
for both PEG (solid
circles) and lysozyme (open circles) cases. The first important
difference between the PEG and lysozyme case occurs at m
2
)
0. Since lysozyme at pH 4.5 is positively charged, we have -
12
) z
P
/2 at m
2
) 0 (with z
p
m
1
, m
2
), where z
P
is the protein
charge.
12,40
For lysozyme, we have previously reported z
p
) 8
( 1.
6
This value corresponds to the protein effective charge
and is 30% lower than the net titration charge,
41
due to
counterion binding.
6
On the other hand, since PEG is a neutral
macromolecule, we have -
12
) 0 at m
2
) 0 (see Figure 3).
We now discuss the dependence of -
12
on salt concentra-
tion. In both cases, -
12
increases with m
2
, consistent with
PEG-KCl net repulsive interactions. However, the increase
observed in the PEG case is significantly larger. To provide a
physical interpretation on the difference in slope between the
two macromolecules, we can use the two-domain model.
2,40
Figure 3. Molality ratios -m2/m1 as a function of PEG molality,
m1, obtained at C2 )0.25 M (solid circles) and C2 )1 M (solid squares)
using the isopiestic method. The solid lines are weighted linear fits
through the data. The preferential interaction coefficients obtained using
ternary diffusion at C2 ) 0.25 M and C2 ) 1 M (open diamonds) are
included for comparison.
Figure 4. Preferential interaction coefficients obtained using ternary
diffusion reported as -12 as a function of salt molality, m2, for the
PEG-KCl-H2O (solid circles) and the lysozyme-KCl-H2O (open
circles) systems. The curves are fits through the data.
PEG-Salt Interactions by Ternary Diffusion J. Phys. Chem. B, Vol. 112, No. 16, 2008 4971
According to this model, the first domain is represented by the
water-osmolyte layers surrounding a macromolecule. This local
domain is in chemical equilibrium with a bulk domain repre-
senting the water-osmolyte remaining solution. Since macro-
molecules interact with the osmolyte and water molecules in
their vicinity, the concentration of osmolyte in the local domain
is different from that of the unperturbed bulk domain. It can be
shown
13,40
that the slope of -
12
(m
2
) is (N
0
/m
0
)(1 - R), where
N
0
is the number of water molecules of the local domain, R is
the osmolyte partitioning constant between the local and the
bulk domains, and m
0
) 55.51 mol kg
-1
. If R < 1, i.e., the
slope of -
12
(m
2
) is positive, the salt concentration of the local
domain is lower than that of the bulk domain. Hence, our results
in Figure 4 show that both PEG and lysozyme are preferentially
hydrated in the presence of KCl. Moreover, we can use the
12
values to calculate N
0
(1 - R) . This quantity represents the
excess of water molecules in the local domain. In Figure 5, we
plot (see N
0
(1 - R) as a function of C
2
for PEG and lysozyme.
We can see that the water excess for PEG is considerably larger
than that for lysozyme, even though the molecular weight of
lysozyme is only 30% lower than that of PEG. This large
difference in water excess can be explained by considering the
structural difference between a protein and polymer. Since
lysozyme is a globular compact protein, only its surface is
accessible to the surrounding fluid. On the other hand, a polymer
molecule behaves like a coil with all its monomeric units
accessible to the surrounding fluid and, therefore, is able to
interact with the salt ions. Furthermore, Figure 5 also shows
that the water excess for the PEG case significantly decreases
as salt concentration increases. This can be related to the
deformability of PEG coils. This observed decrease in water
excess is also consistent with a disruption of structured water
surrounding the PEG macromolecules induced by the presence
of salt ions. As C
2
increases, the polymer coil collapses on itself
due to PEG-KCl net repulsive interactions. Clearly, this result
is consistent with the salt-dependence of (D
11
)
V
(/
0
) shown
in Figure 2.
Examination of Cross-Diffusion Coefficients. We will now
discuss the behavior of the cross-diffusion coefficients in relation
to the main diffusion coefficients and the preferential interaction
coefficients. We will consider the solvent-frame diffusion
coefficients, due to their direct relation to chemical potential
derivatives (see eq 5a-d). The difference between the volume-
frame and solvent-frame values is small and marginal for the
interpretations reported below. We will specifically address the
behavior of the ratios: (D
12
)
0
/(D
11
)
0
and (D
21
)
0
/(D
22
)
0
. This
allows us to remove the explicit dependence of the cross-
diffusion coefficients on the tracer diffusion coefficients of
macromolecule and salt.
The two diffusion-coefficient ratios can be expressed in terms
of the chemical potential derivatives,
ij
(c)
, using eq 5a-d:
If (L
12
)
0
) 0, the ratios: (D
12
)
0
/(D
11
)
0
and (D
21
)
0
/(D
22
)
0
become
equal to the thermodynamic ratios:
12
(c)
/
11
(c)
and
21
(c)
/
22
(c)
,
respectively. Indeed the Nernst-Hartley equations,
15
which are
valid at infinite dilution, predict (L
12
)
0
< 0 for charged
macromolecules, and (L
12
)
0
f 0 as the charge on the macro-
molecule becomes zero. This limiting case applies to PEG. We
will now examine if Onsager cross-diffusion coefficients can
be indeed neglected for neutral macromolecules by numerically
comparing our diffusion and the thermodynamic ratios. The
corresponding values are reported in Table 2. We can see that
(D
12
)
0
/(D
11
)
0
is 1 order of magnitude lower than
12
(c)
/
11
(c)
within
our salt concentration range. Although we have assumed
11
(c)
)
RT/C
1
, this approximation is not expected to significantly affect
the observed large discrepancy. We therefore conclude that the
assumption (L
12
)
0
) 0 in eq 11a cannot be applied for the
interpretation of (D
12
)
0
/(D
11
)
0
. On the other hand, we can see
in Table 2 that the discrepancy between (D
21
)
0
/(D
22
)
0
and
21
(c)
/

22
(c)
is only 1-2%. Thus (L
12
)
0
does not play an important role
in eq 11b, and (D
21
)
0
/(D
22
)
0
is essentially a thermodynamic
quantity as shown by eq 10. The nearly equivalence between
(D
21
)
0
/(D
22
)
0
and
21
(c)
/
22
(c)
can be understood by considering the
diffusion experiment for which an initial sharp boundary exists
between two solutions having same salt concentration and
different macromolecule concentrations. Since macromolecules
diffuse relatively slowly, the fast-diffusing salt component (and
the solvent) will reach a quasi-equilibrium condition, (J
2
)
0

0, in the presence of a slowly dissipating concentration gradient
of macromolecules. According to eq 1b, this implies that
(D
21
)
0
C
1
- (D
22
)
0
C
2
or (D
21
)
0
/(D
22
)
0
-(C
2
/C
1
)
2
)

21
(c)
/
22
(c)
. Clearly, this interpretation cannot be extended to the
opposite diffusion experiment with the two solutions having
same macromolecule concentration and different salt concentra-
tion. This is because a salt concentration gradient will dissipate
relatively fast. Thus (D
12
)
0
/(D
11
)
0
is expected to significantly
depend on kinetic parameters. For ternary systems, with two
solutes of different size, we can therefore conclude that only
the salt diffusion ratio, (D
21
)
0
/(D
22
)
0
, can be interpreted merely
using thermodynamic quantities. On the other hand, the
contribution of (L
12
)
0
cannot be neglected for the interpretation
of the macromolecule diffusion ratio, (D
12
)
0
/(D
11
)
0
, even for
neutral macromolecules.
We now compare our results on (D
12
)
0
/(D
11
)
0
and (D
21
)
0
/
(D
22
)
0
for the PEG case with those previously obtained for the
Figure 5. Water excess, N0(1 - R), a functions of KCl concentration,
C2, for the PEG-KCl-H2O (solid circles) and the lysozyme-KCl-
H2O (open circles) systems. The curves are fits through the data. The
values of N0(1 - R) for the PEG case were calculated using (-12)-
(m0/m2). The values of N0(1 - R) for the lysozyme case were calculated
using (-12 - zP/2)(m0/m2) with zP ) 8.
TABLE 2: Comparison between (D
ij
)
0
/(D
ii
)
0
and
ij
(c)
/
ii
(c)
Ratios
Ch2 (M) 0.20 0.25 0.30 0.50 1.00 2.00 2.30
(D12)0/(D11)0 0.0012 0.0011 0.0011 0.0011 0.0011 0.0011 0.0011

12
(c)
/
11
(c)
0.0114 0.0106 0.0109 0.0094 0.0085 0.0081 0.0082
(D21)0/(D22)0 8.1 9.8 11.9 18.0 33.4 61.7 70.1

21
(c)
/
22
(c)
8.3 9.9 12.1 18.3 33.8 62.5 71.0
(D
12
)
0
(D
11
)
0
)

12
(c)
+ [(L
12
)
0
/(L
11
)
0
]
22
(c)

11
(c)
+ [(L
12
)
0
/(L
11
)
0
]
21
(c)
(11a)
(D
21
)
0
(D
22
)
0
)

21
(c)
+ [(L
12
)
0
/(L
22
)
0
]
11
(c)

22
(c)
+ [(L
12
)
0
/(L
22
)
0
]
12
(c)
(11b)
4972 J. Phys. Chem. B, Vol. 112, No. 16, 2008 Tan et al.
lysozyme case.
6
We note that, in the limit of C
1
f 0, the
denominator of eq 11a approaches
11
(c)
) 1/C
1
and (D
12
)
0
/(D
11
)
0
becomes directly proportional to C
1
. To remove this expected
C
1
dependence, the comparison between these two macromol-
ecules is performed by examining [(D
12
)
0
/(D
11
)
0
]/C
1
as a
function of C
2
. In Figure 6, we can see that the values
corresponding to the PEG case are about 4-fold larger than those
related to the lysozyme case. Although the ratio (D
12
)
0
/(D
11
)
0
significantly depends on both thermodynamic and transport
parameters, the difference in [(D
12
)
0
/(D
11
)
0
]/C
1
between the two
macromolecules implies that PEG affinity for water and,
consequently, its tendency to migrate toward lower salt con-
centration are higher than those of the lysozyme. This is
consistent with our preferential hydration results in Figure 5.
In Figure 7, we show the (D
21
)
0
/(D
22
)
0
values for the PEG
and lysozyme cases as a function of C
2
. The corresponding
curves increase with salt concentrations and display a behavior
similar to that shown in Figure 4 for -
12
. This is consistent
with eq 10, which shows that the values of the (D
21
)
0
/(D
22
)
0
slopes are systematically higher than the corresponding values
of -
12
, due to the positive contribution of C
2
Vh
1
.
Conclusions
We have reported ternary diffusion coefficients for the PEG-
KCl-water system. First, the experimental comparison between
diffusion and isopiestic measurements has allowed us to
demonstrate that multicomponent diffusion coefficients can be
used to determine preferential interaction coefficients. This result
is not only important for the examination of the accuracy of
the thermodynamic data obtained using ternary diffusion but it
also represents a significant contribution to the validation of
three postulates of nonequilibrium thermodynamics applied to
diffusion processes involving macromolecules. Second, we have
observed a large preferential hydration of the PEG coils in the
presence of KCl. This hydration was found to be about 1 order
of magnitude larger than that observed for lysozyme, a protein
of similar molecular weight. Third, the diffusion coefficient
ratios: (D
12
)
0
/(D
11
)
0
and (D
21
)
0
/(D
22
)
0
, which describe the
magnitude of cross-diffusion coefficients, are larger for the
PEG-KCl-water system than those for the lysozyme-KCl-
water system at the same solute concentrations. This difference
can be related to the corresponding difference in preferential
hydration. We have also established that only the diffusion ratio
(D
21
)
0
/(D
22
)
0
is essentially a thermodynamic quantity. The other
diffusion ratios strongly depend on the Onsager cross-diffusion
coefficient, (L
12
)
0
, even for neutral macromolecules.
Acknowledgment. This work was supported by the ACS
Petroleum Research Fund (47244-G4) and TCU Research and
Creative Activity Funds.
Supporting Information Available: Experimental details
related to individual ternary diffusion experiments and binary
experiments on the KCl-water system. This material is available
free of charge via the Internet at http://pubs.acs.org.
References and Notes
(1) Timasheff, S. N. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 7363-
7367.
(2) Parsegian, V. A.; Rand, R. P.; Rau. D. C. Proc. Natl. Acad. Sci.
U.S.A. 2000, 97, 3987-3992.
(3) Timasheff, S. N. Proc. Natl. Acad. Sci. U.S.A. 2002, 99, 9721-
9726.
(4) Schellman, J. A. Q. ReV. Biophys. 2005, 38, 351-361.
(5) Wyman, J., Jr. AdV. Protein Chem. 1964, 19, 223-286.
(6) Annunziata, O.; Paduano, L.; Pearlstein, A. J.; Miller, D. G.;
Albright, J. G. J. Phys. Chem. B 2006, 110, 1405-1415.
(7) Arakawa, T.; Bhat, R.; Timasheff, S. N. Biochemistry 1990, 29,
1914-1923.
(8) Courtneay, E. S.; Capp, M. W.; Anderson, C. F.; Record, M. T.,
Jr. Biochemistry 2000, 39, 4455-4471.
(9) Leaist, D. G. J. Phys. Chem. 1989, 93, 474-479.
(10) Albright, J. G.; Annunziata, O.; Miller, D. G.; Paduano, L.;
Pearlstein, A. J. J. Am. Chem. Soc. 1999, 121, 3256-3266.
(11) Annunziata, O.; Paduano, L.; Pearlstein, A. J.; Miller, D. G.;
Albright, J. G. J. Am. Chem. Soc. 2000, 122, 5916-5928.
(12) Annunziata, O.; Paduano, L.; Albright, J. G. J. Phys. Chem. B 2006,
110, 16139-16147.
(13) Annunziata, O.; Paduano, L.; Albright, J. G. J. Phys. Chem. B 2007,
111, 10591-10598.
(14) We use the subscript 1 for the macromolecule and the subscript
2 for the osmolyte, consistent with the literature on multicomponent
diffusion. However, the subscript 2 for the macromolecule and the
subscript 3 for the osmolyte are extensively used in relation to the
thermodynamic description of protein mixtures.
(15) Gosting, L. J. AdV. Protein Chem. 1956, 11, 429-554.
(16) Paduano, L.; Annunziata, O.; Pearlstein, A. J.; Miller, D. G.;
Albright, J. G. J. Cryst. Growth 2001, 232, 273-284.
(17) Albertsson, P. A. Partition of Cell Particles and Macromolecules;
Wiley: New York, 1986.
(18) McPherson, A. Crystallization of Biological Macromolecules; Cold
Spring Harbor: New York, 1998.
(19) Edlund, U.; Albertsson, A. C. AdV. Polym. Sci. 2002, 157, 67-
112.
(20) Vergara, A.; Annunziata, O.; Paduano, L.; Miller, D. G.; Albright,
J. G.; Sartorio, R. J. Phys. Chem. B 2004, 108, 2764-2772.
(21) Capuano, F.; Vergara, A.; Paduano, L.; Annunziata, O.; Sartorio,
R. J. Phys. Chem. B 2003, 107, 12363-12369.
(22) Fitts, D. D. Nonequilibrium Thermodynamics: A Phenomenological
Theory of IrreVersible Processes in FLuid Systems; McGraw-Hill: New
York, 1962.
(23) Albright, J. G.; Rizzo, P.; Rard, J. A. J. Chem. Thermodyn. 1998,
30, 327-352.
(24) Kirkwood, J. G.; Baldwin, R. L.; Dunlop, P. J.; Gosting, L. J.;
Kegeles, G. J. Chem. Phys. 1960, 33, 1505-1513.
Figure 6. The ratio [(D12)0/(D11)0]/C1 as a function of salt concentration,
C2, for the PEG-KCl-H2O (solid circles) and the lysozyme-KCl-
H2O (open circles) systems. The curves are linear fits through the data.
Figure 7. The ratio (D21)0/(D22)0 as a function of salt concentration,
C2, for the PEG-KCl-H2O (solid circles) and the lysozyme-KCl-
H2O (open circles) systems. The curves are linear fits through the data.
PEG-Salt Interactions by Ternary Diffusion J. Phys. Chem. B, Vol. 112, No. 16, 2008 4973
(25) Miller, D. G.; Vitagliano, V.; Sartorio, R. J. Phys. Chem. 1986,
90, 1509-1519.
(26) Dunlop, P. J.; Gosting, L. J. J. Phys. Chem. 1959, 63, 86-93.
(27) Woolf, L. A.; Miller, D. G.; Gosting, L. J. J. Am. Chem. Soc. 1962,
84, 317-331.
(28) Onsager, L. Phys. ReV. 1931, 38, 2265-2279.
(29) Miller, D. G. J. Phys. Chem. 1959, 63, 570-578.
(30) Vergara, A.; Paduano, L.; Sartorio, R.; Vitagliano, V. J. Phys. Chem.
B 1999, 103, 8732-8738.
(31) Miller, D. G.; Albright, J. G. In Measurement of the Transport
Properties of Fluids: Experimental Thermodynamics; Wakeham, W. A.,
Nagashima, A., Sengers, J. V., Eds.; Blackwell Scientific Publications:
Oxford, 1991; Vol. 3, pp 272-294.
(32) Miller, D. G. J. Phys. Chem. 1988, 92, 4222-4226.
(33) Rard, A. J.; Miller, D. G. J. Chem. Eng. Data 1980, 25, 211-215.
(34) Annunziata, O.; Buzatu, D.; Albright, J. G. Langmuir 2005, 21,
12085-12089.
(35) Tanford, C. Physical Chemistry of Macromolecules; Wiley: New
York, 1961.
(36) Vergara, A.; Paduano, L.; Sartorio, R. J. Polym. Sci., Part B: Polym.
Phys. 2002, 40, 43-51.
(37) Zhang, H.-L.; Han, S.-J. J. Chem. Eng. Data 1996, 41, 516-520.
(38) Taboada, M. E.; Galleguillos, H. R.; Graber, T. A.; Bolado, S. J.
Chem. Eng. Data 2005, 41, 264-269.
(39) Sridharan, S.; Cukier, R. I. J. Phys. Chem. 1984, 88, 1237-1242.
(40) Record, M. T., Jr.; Anderson, C. F. Biophys. J. 1995, 68, 786-
794.
(41) Tanford, C.; Wagner, M. L. J. Am. Chem. Soc. 1954, 76, 3331-
3336.
4974 J. Phys. Chem. B, Vol. 112, No. 16, 2008 Tan et al.